Final Study Protocol and Analytical Plan...1 Final Study Protocol and Analytical Plan 2...

Final Study Protocol and Analytical Plan 1

Principal/Investigator: Margarita Alegria 2

Protocol Title: Effectiveness of DECIDE in Patient-Provider Communication, Therapeutic Alliance and Care 3 Continuation 4

Funding: PCORI - Patient Centered Outcomes Research Institute 5

I. Background and Significance 6

I.A. Historical Background. Most people with mental disorders in the United States are untreated or poorly 7 treated.1 Numerous researchers, government agencies, and advocates1-3 call for interventions to enhance treatment 8 initiation and quality. One way to improve quality and retention in mental health care is to implement shared 9 decision making (SDM). When patients and providers engage in SDM, patients’ preferences are taken into 10 consideration for their treatment, resulting in more appropriate care, increased satisfaction, and ideally, better health 11 outcomes.4,5 Examining SDM in behavioral health care among populations in safety net settings is an urgent need 12 given that the Federal Patient Protection and Affordable Care Act (ACA) will make an estimated 32 million 13 uninsured individuals, mostly minorities, eligible for Medicaid coverage.6 Roughly 20-30% of this population 14 suffers from a behavioral health disorder.7 15

A major obstacle to SDM for patients of color is inadequate provider appreciation of minority patients’ preferences 16 for interpersonal relations.8, 9 Providers rarely receive training in how to motivate minority patients to voice their 17 treatment concerns or preferences. Providers display fewer patient-centered behaviors,10 are less receptive to 18 question asking, and tend to demonstrate greater verbal dominance11 with minorities than with white patients. These 19 actions often result in misunderstandings, inadequate services, and failed treatment alliances.12 Minority patients 20 may infer prejudice or perceive a negative attitude from their provider, thus reducing the likelihood that they 21 perceive receiving quality care.13 Clinicians face new demands connecting with patients with different customs, 22 values and experiences, and addressing these challenges will likely improve SDM and patient-centered quality care. 23 Language barriers can also be detrimental;14 patients who do not speak the same language as their providers report 24 worse outcomes15 and higher dropout rates.16-18 Tackling these barriers requires new, innovative interventions at the 25 provider and patient level. 26

I. B. Previous clinical studies leading up to and supporting the proposed research. In this proposal, we test the 27 effectiveness of the DECIDE-PA+PC intervention among non-Latino white, African-American, Latino, and Asian 28 patients in community health centers. Our research team’s pilot and randomized trial found that a patient 29 intervention (DECIDE-PA) improved activation and self-management in mental health care.19-22 However, minority 30 patients expressed concern that becoming ‘activated’ threatened the relationships they had developed with their 31 providers. This feedback meshed with prior studies showing that providers working under strict time constraints and 32 immediate treatment priorities may be more directive and limit patient-initiated talk.23 In the DECIDE-PC 33 intervention described in this protocol, we intended to tackle these barriers. We hypothesized that adding a provider 34 component (DECIDE-PC) to the patient component (DECIDE-PA) would enhance providers’ receptivity to 35 patients’ activation and self-management, improve therapeutic alliance and communication, and improve SDM and 36 perceived quality of care. 37

The DECIDE-PA intervention (previously called RQP-MH) had its roots in a community-based social action 38 intervention, the Right Question Project (RQP), which was designed to empower participants in social and health 39 situations that required decision making. The DECIDE-PA intervention was a product of the collaboration between 40 the non-profit group which created RQP (Rothstein and Santana), patients, clinicians, and administrators in mental 41 health agencies, and the Disparities Research Unit (DRU), MGH team, which was previously the Center for 42 Multicultural Mental Health Research (CMMHR) at Cambridge Health Alliance. Considerations of cultural, socio-43 economic, and clinical factors for patients were gleaned from stakeholder groups in the adaption of RQP to 44 DECIDE-PA (see Polo et al. 2012 for details).12 The continued engagement of patients, clinicians, and 45 administrators in the DECIDE-PA+PC study was vital to ensure that the intervention continued to be relevant and 46 meaningful for patients and to attain sustained improvements in patient care. 47

Downloaded From: https://jamanetwork.com/ by a Non-Human Traffic (NHT) User on 01/14/2021

The current study was guided by principles of community-based research and was committed to ensure patients, 48 clinicians, and clinic administrators had a purposeful voice in study design and implementation and dissemination of 49 findings. By incorporating the diverse skills, knowledge and expertise of patients, providers and clinic 50 administrators, the research was more likely to be useful and relevant to community members.24 The research PI 51 and team has an extensive and well-respected history of successful collaboration with patients, clinicians, 52 administrators, and other stakeholders in innovative mental health services studies that aim to improve the lives of 53 multicultural populations. The unique collaborative nature of the research center was evident in the conduct of the 54 DECIDE-PA study. We continued to respect the voice of patients and stakeholders by incorporating extensive input 55 in design, implementation, and dissemination. 56

I.C. Rationale behind the current research and potential benefits to patients and/or society. Our study is one of 57 the first to test whether changes in patient activation and self-management together with a provider coaching 58 program to increase provider receptivity improves SDM and patient’s perception of behavioral health care quality. 59 Although past research suggests that racial/ethnic minority patients take a more passive role in treatment and are less 60 likely to discuss information with a health care provider, the DECIDE-PA study demonstrated that racial/ethnic 61 minority patients’ level of activation and self-management can increase. However, the main tenet of SDM is that 62 two active participants, the patient and the provider, are needed.25 63

This study also fills a gap for scientifically rigorous research in clinics that ethnic/racial populations depend on to 64 receive mental health care services.26 Research shows that minority patients do not have equal access to high quality 65 care.27, 28 Administrators and providers are eager to implement interventions but first need strong evidence of 66 improved quality or outcomes in resource-constrained safety net environments.29 The collaborative engagement of 67 patients, clinicians, and administrators helps ensure that DECIDE PA+PC is relevant and meets their needs. Further, 68 the study was designed to triangulate data from multiple perspectives (i.e., patient, clinician, and independent 69 observer) to allow for better measurement of SDM and other outcomes. 70

STUDY AIMS AND PROCEDURES 71

We include below the study aims and procedures that reflect our completed protocol. 72

II. Specific Aims 73

II.A. Specific Aims: Aim 1: Test the effectiveness of the DECIDE PA+PC intervention and the marginal benefit 74 of DECIDE-PA or PC compared to usual care in improving shared decision making and patient-perceived 75 quality of behavioral health care. 76 Aim 2: Test whether patient-centered communication and therapeutic alliance mediate the effect of the 77 DECIDE PA+PC intervention on shared-decision making. 78 Aim 3: Explore whether ethnic/racial or language matching moderates the relationship between the effect of the 79 DECIDE PA+PC intervention on shared decision making, quality of behavioral health care, patient-centered 80 communication and therapeutic alliance. 81 82

III. Subject Selection 83 III.A.1. Provider inclusion criteria: Provider participants 84 in this study consisted of regular paid staff members that 85 provide behavioral health services (i.e., psychotherapy 86 and/or medication) to adult outpatients at each of the 87 participating study clinics. No other criteria were required. 88 Across all study sites, approximately 80 providers were 89 ultimately targeted for recruitment to participate in the 90 study. The final number of participating providers was 79. 91 At each site (a “site” may consist of multiple clinics within 92 one hospital) approximately 4 -10 providers took part in the 93 intervention trial. 94 95 III.A.2. Patient inclusion criteria: Enrollment was limited 96 to patients ages 18-80, non-Latino White or Latino, Black, 97


or Asian, who were receiving mental health treatment at one of the collaborating clinics, from a participating 98 provider. Non-English speaking patients were included in the study (i.e., patients who speak Spanish or Mandarin). 99

Across all study sites, 360 patients were targeted to be randomized into the control and intervention arms of the 100 study, to achieve the target sample size of approximately 300 given expected attrition of 20%. We finalized the 101 study sample with 312 patients. We refer to these patients as RCT patients. We anticipated enrolling approximately 102 48-72 patients at each site, half of which would go into the control arm and the other half into the intervention arm 103 of the study. Additionally, we planned to recruit approximately 124 patients to consent to having one clinical session 104 audio recorded with their respective consented provider, to use to help train providers enrolled in the intervention 105 arm of the study. These patients were not randomized to the control or intervention arms. We refer to these patients 106 as NRCT. We finalized the study sample with 101 NRCT patients. 107

III.A.3. Patient exclusion criteria: Patients were excluded for the following: positive screen to a diagnosis of 108 Bipolar Disorder or taking medication for Bipolar Disorder (i.e. Lithium), diagnosis of Schizophrenia; endorsed 109 active suicidal ideation in the last 30 days or pending hospitalization; participation in previous DECIDE trials, 110 failure to pass the mini-cognition assessment for participants 65 years of age or older; or indication by the provider 111 that a patient was too sick to participate in the study. Patients younger than 18 years of age or older than 80 years of 112 age were also excluded. Suicidal patients or patients who were pending hospitalization in the coming month had the 113 option to be rescreened one month following their initial screening if these previous conditions no longer applied. 114 Suicidal patients were referred for immediate help following a study emergency protocol determined in 115 collaboration with clinic staff. 116

III.A.4. Focus group inclusion criteria: Patients and providers who completed their participation in the study were 117 eligible to participate in focus groups following the clinical trial. Patients and providers who did not participate in 118 the study were also invited to participate, as described further below. A total of 30 patients and 19 providers 119 participated in our focus groups (49 total). 120

III.B. Source of Subjects and Recruitment Methods: We recruited at selected community health clinics, both 121 those where we had established collaborations, as well as new partnerships that we forged. The clinics were chosen 122 based on criteria including patient and provider volume, demographics in terms of a high percentage of Latino, 123 Black, Asian and non-Latino White patients, previous collaborations, and skilled Site Leaders/Co-Investigators. We 124 recruited providers who are regular staff at each clinic. Only patients of these providers were eligible for 125 randomization. 126

A series of presentations and meetings were held to introduce staff at each of the clinics to the study. We collected 127 informed consent forms from those providers that were interested in participating (as was done in DECIDE-PA). To 128 screen patients, we enlisted the help of providers as well as administrative assistants (AA) at each of the clinics, who 129 had access to provider caseloads, schedules, and patient demographics. To coordinate recruitment, we asked AAs to 130 select those who patients who met basic eligibility criteria: between 18-80 years old, non-Latino White or Latino, 131 Black, or Asian, and not at risk for self-harm if known to the clinic. At selected clinics we had access to EPIC 132 electronic records to help with this process. Patients were assured that accepting or declining to participate would 133 not affect their standard clinical care. 134

IV. Subject Enrollment 135

IV. A.1. Methods of enrollment, including procedures for patient registration and/or randomization: The 136 DECIDE PA+PC intervention involved a randomized controlled trial of patients within providers participating in the 137 proposed study. The original target was 8-10 providers (always in pairs at participating clinics, half going to the 138 intervention and half to the control condition) every 6 months. Approximately 4-8 patients from each of these 139 providers participated in the study design. In some clinics this number varied depending on patient flow. 140

At each clinic, participating providers completed a baseline interview, the RA1, after which they were randomized 141 to the control or intervention (DECIDE-PC) arms. All providers audio recorded one clinical session with 1-2 142 separate patients (called NRCT patients) in order to help train providers enrolled in the intervention arm of the 143 study. Providers in the control arm continued administering usual care. Providers randomized to the intervention arm 144 participated in a 1.5 day training led by coaches as well as 1-6 follow up coaching calls throughout the study period. 145

Figure 1:

Patient

Recruitment/

Screening


Following provider training, consented patients were recruited and randomly assigned to DECIDE-PA or usual care. 146 We implemented a randomized clinical trial, whereby patients were randomized to the DECIDE-PA intervention or 147 usual care within each of the clinics and with each of the randomized consenting providers receiving DECIDE-PC or 148 not. Patients assigned to the intervention received 1-3 training sessions delivered by a Care Manager (CM) over a 149 period of 2-6 months; patients in the control arm received treatment as usual. Patients in the intervention arm elected 150 to complete their trainings either in person (at the participating clinic or in their homes), or over the phone. Providers 151 and participants in the control arm of the study were eligible to receive the DECIDE-PC and DECIDE-PA 152 interventions respectively, once their participation in the study was complete. 153

Assessment of outcomes: Outcomes were measured by research assistants (RAs) who were not involved in the 154 provider-or patient-level trainings. Patients were assessed at 3 time points during the study, with the RA1 or baseline 155 assessment completed at the time of their enrollment in the study and before treatment exposure. The second 156 assessment, the RA2, occurred 1 to 2 months following the first, and took place within 24 hours of the 157 patient/provider clinical session recording. This was planned to be done after both the provider and patient had 158 treatment exposure. However, in some cases patients in the intervention were given RA2, without having treatment 159 exposure. Patients who lost their health insurance prior to the clinical session being recorded, but who wished to 160 remain in the study, had the opportunity to have one clinical session with their provider paid for by the study. The 161 third and final assessment, the RA3, occurred 3 to 6 months after the baseline assessment and signaled the 162 conclusion of the patient’s participation in the study. Patients could elect to complete these assessments either in 163 person (at the participating clinic or in their homes) or over the phone. Providers completed an online assessment, 164 the RA2, ideally within 24 hours of recording a clinical session with their participating patients. Providers 165 randomized to the intervention participated in follow-up calls with coaches, one call for every session recorded, as 166 well as a final wrap-up call. Once all calls were completed, providers completed the final assessment of the study. 167

We summarize here the three different methods of data collection: 168

1) Patient assessments were collected in face-to-face interviews or by phone at baseline (RA1), 1-2 months 169 (RA2), and 3 to 6 months (RA3). Follow up interviews were conducted with patients even if the patient were no 170 longer receiving care at the clinic or no longer was seeing their study provider. 171

2) Provider assessments were collected in face-to-face interviews or by phone. Providers were asked about their 172 general interactions with patients at baseline and post-intervention, and specifically about their clinical encounter 173 after the clinical visit (RA2). 174

3) Coded audio recordings of the clinical visit (RA2 only). We obtained audio recordings of each clinical visit 175 for the participating patients for both intervention and control providers. These recordings were coded by blinded 176 research staff for SDM, patient-centered communication, therapeutic alliance, provider receptivity, and global 177 interaction rating. 178

Difficult to reach patients: There were often difficult to reach patients. As the study progressed, we ramped up 179 follow up procedures to assist with patient retention in the study. We offered more flexibility to the patients in terms 180 of when and where the interviews could be completed. We completed interviews after hours, during the weekends, 181 and at times in the patient’s home. Our call protocol was increased to 6 daily attempts: 2 in the morning, 2 in the 182 afternoon, and 2 in the evenings. In some cases patients’ phones were disconnected, their voicemails were full, or 183 we had outdated information in our records. We worked with clinic administrative staff to obtain updated contact 184 information. For patients who we still could not reach, we sent them a letter informing them that we would visit their 185 homes to check on their well-being and complete the assessments. The research assistants were all trained in safety 186 procedures for home visits. Permission to visit patients in their homes was included in the consent forms. 187

Focus groups: Towards the end of the study period, two community forums (breakfast meetings) and 7 focus 188 groups were held. These forums and groups were used not only to disseminate preliminary results from the study, 189 but also to enable us to receive feedback and insight from participants in order to inform future research. Patient 190 focus groups took place in English, Spanish, and Mandarin and were audio recorded for quality control and 191 transcription purposes. 192

193 IV. A.2. Recruitment and Consent Procedures 194


Providers: Research staff worked together with Site Leaders at each participating clinic to recruit providers who 195 were regular staff members of the clinic. Dr. Margarita Alegria and the research team then conducted a series of 196 presentations and meetings to introduce the study to providers and clinic staff. Following the presentations, the 197 research staff visited clinics to answer any additional questions and to administer informed consent with providers 198 who expressed interest in participating. Provider Informed Consent forms were explained during the individual and 199 group meetings with providers, including information about the potential risks and benefits of enrollment, the 200 collection of provider data, the need for audio recordings of patient visits, the voluntary nature of their participation, 201 and the option of withdrawal at any point during the course of the study. 202 203 Patients: In terms of patient enrollment, study Research Assistants (RAs) collaborated with the participating 204 administrative staff at each clinic in order to obtain the clinical schedule for enrolled providers. The clinical 205 schedules were then used to determine which patients were eligible to approach for participating providers. Patients 206 were approached if they fit the age requirement (age confirmed by administrative staff) and if the patient did not 207 display severe levels of cognitive impairment (judged by the RA or as determined by provider in some rare cases 208 [n=9]). The RAs approached patients that arrived at the clinic well in advance (at least 15 minutes) of their 209 appointment, or as patients were completing their appointments. We also posted flyers in Spanish, English, and 210 Mandarin at participating clinics to increase our equitable recruitment at all clinics. At certain clinics, RAs were 211 granted limited access to electronic scheduling systems (e.g., EPIC), to determine provider schedules or enrolled 212 patients’ next appointments. Patient eligibility and consent to participate was then determined during an in person 213 screening visit. All participation in the study was voluntary. Patients were assured that accepting or declining to 214 participate would not affect their clinical care, and their participation in the study was completely voluntary. RAs 215 noted patients who declined participation with a study ID so as not to approach them in the future and for tracking 216 purposes. 217 218 Patients were given as much time as they needed to think about participating. Patients who might have been eligible 219 for the study, but who did not have time to speak with an RA in the clinic originally signed a “pre-consent” form 220 giving permission to the study staff to contact them by email or phone. The RA contacted them with more 221 information about the study and to set up a time to administer the screening. Once the research team moved to 222 Massachusetts General Hospital, the hospital’s IRB no longer required a “pre-consent” form. In lieu of the “pre-223 consent” form, RAs provided interested patients a study information sheet that included the research team’s contact 224 information. RAs then collected patient information during informed consent and screening, after the patient had 225 contacted the research team. 226

RAs took patients to a separate, private room in order to review the informed consent form and perform the screener 227 for eligibility. Patients over the age of 65 were further screened for cognitive impairment using a mini-cognition 228 assessment to ensure they were capable of participation. Patients in substance abuse clinics only were asked to 229 demonstrate capacity-to-consent by answering 8 out of 10 questions about the consent form correctly in order to 230 participate in the study, given concern about intoxication or use of substances prior to the consent process. Those 231 who could not complete the capacity to consent measure were considered ineligible, even if their screener indicated 232 eligibility. Patients were informed that the research team would make every effort to ensure that the information 233 they shared during the course of the trainings and assessments was kept confidential and only reported in aggregate 234 form. 235

English and non-English speaking patients were also included in the study (i.e., patients who spoke Spanish or 236 Mandarin). The study was conducted fully in three languages: English, Spanish, or Mandarin Chinese). To assist 237 with their equitable recruitment, RAs in charge of recruitment were fluent in at least two of the three languages to 238 assist and be able to fully consent non-English speaking patients. Consent forms were professionally translated and 239 available in all three languages. 240 241 Focus Groups: Patients and providers who previously participated in the study were contacted by an RA to gauge 242 interest in participating in a focus group. Focus Group Informed Consent was obtained verbally, since no PHI was 243 collected during the course of the groups. 244 245 Provider final qualitative data: As we finalized the study, we examined potential challenges and barriers to 246 implement our intervention in real world practice. We collected qualitative data from a portion of the providers (n= 247 41) on clinical and organizational challenges and facilitators to implementing the DECIDE-PC provider intervention 248


in a real-world setting. We used a survey questionnaire with 6 questions. Participants answered the survey on 249 RedCap. The entire survey took less than 15 minutes of their time. Participants received a $50 gift card upon 250 completion of the survey. 251

252 IV.A.3. Remuneration 253

Consented patients who were screened for the study, but who did not meet eligibility requirements based on the 254 screener, were not included in the study, but received a $10 gift card. 255

The 1-2 NRCT patients per provider, who consented to having their clinical session with their provider audio 256 recorded, completed a brief baseline assessment, the RA0, (i.e., the patient is not randomized to the control arm or 257 intervention arm of the study) and received a $20 gift card. 258

Patient participants in the intervention arm of the study were given a total of $120 worth of gift cards for their 259 participation. This includes compensation for 3 interviews ($25 for each of the first two assessments and $40 for the 260 final assessment) and 3 training sessions ($10 per session to help with transportation and child care expenses to 261 participate in the trainings). At the conclusion of the study, we increased the incentive from $25 to $50 for the 262 clinical recording and the 2nd interview, to facilitate patient completion of study protocol. 263

Patient participants in the control group received a total compensation of $90 for the 3 interviews ($25 for each 264 of the first two assessments and $40 for the final assessment). At the conclusion of the study, we increased the 265 incentive from $25 to $50 for the clinical recording and the 2nd interview. 266

Provider participants in the intervention arm of the study received incentives for the DECIDE-PC behavioral 267 health provider trainings ($300 per provider for participating in the training) and received continuing education 268 credits for their respective disciplines (e.g., social work, psychology, psychiatry). Providers received up to 18.5 total 269 continuing education credits (11.5 for the training and up to 1 additional credit for each follow-up coaching call and 270 1 for the wrap up call.) Providers were paid $50 for each of two research assessments (total $100), conducted by the 271 RAs. We also paid $50 to providers per patient to take part in a self-administered post-appointment assessment 272 (RA2) that helped them evaluate how they were doing with their patients. Therefore, providers in the intervention 273 group (i.e., those that received the DECIDE training) received a total of $700 [($300 for provider training) + ($50 274 per research assessment *2 assessments [baseline (RA1 and RA3]) + ($50 per patient post-appointment, self-275 administered assessment RA2 * up to 6 patients)] to participate in the study. 276

Provider participants in the control group (i.e., did not receive the DECIDE training, but participated in all other 277 aspects of the study) received a total of $400 [($50 per research interview *2 interviews) + ($50 per patient post-278 appointment, self-administered survey * 6 patients)] to participate in the study. 279

Patient and Provider participants in the focus groups received a $50 gift card for their participation. 280

V. Interventions 281

282 V.A.1. Provider Intervention: Provider coaching focuses on augmenting patient-centered communication and 283 therapeutic alliance as a possible underlying pathway by which SDM can take place. Provider Coaching targets three 284 areas that were identified in our previous Patient Provider Encounter Study30-32 as problematic in forming good 285 provider-patient interactions as well as using recommended coaching on patient-centered communication shown to 286 be effective in clinical encounters.33-39 These are: 1) lack of perspective taking or the ability to step outside of one’s 287 own experience and accurately identify the emotions and perceptions of others;40 2) frequent attributional errors 288 that involve dispositional inferences,41 where one attributes negative behaviors of out-group members (people of 289 different ethnicity/race or language) to innate traits whereas negative behavior of in-group members is attributed to 290 more situational factors as well as inaccurate identification of patient’s feelings and emotions; and 3) decreased 291 receptivity to patient participation and collaboration in decision making. In addition, we included three 292 additional areas reported in the literature: 4) patient activation; 5) patient engagement; 6) global impressions (i.e. 293 warmth, respect for patient); and 7) encouraging open communication. 294


The training consisted of three parts totaling ~20 hours. The first part included 12 hours of a small group 295 experiential workshop, including two hours of individual feedback on the seven targeted areas of intervention 296 mentioned above. The second part included 6 hours of individual coaching. A third part consisted of a 1 hour wrap-297 up call to summarize the coaching work. 298

Part 1: Provider receptivity to SDM was introduced in a group workshop. Providers were given an overview of the 299 goals and logistics of the trainings, followed by a brief presentation of the research behind patient activation and our 300 own study findings. We explained how the provider intervention teaches communication skills in listening, eliciting 301 the patient’s agenda, encouraging question asking, and illness management education. Videos of two contrasting 302 interviews (responsive and non-responsive providers) were presented and discussed, focusing on attentiveness (how 303 patients’ concerns and understandings are taken seriously by the provider), facilitation (encouraging patients to 304 express concerns in their own words and facilitating self-management and activation) and collaboration (supporting 305 patients as partners in the process of mental health care). Techniques demonstrated included “giving the floor” to the 306 patient (attentiveness), focusing on the voice of the patient rather than the voice of medicine (facilitation), and 307 validating the patient as a co-producer/partner of treatment outcomes (collaboration). The training emphasized that 308 allowing patient-initiated topics signals to patients that they are responsible for their treatment. The non-responsive 309 interview exemplified non-specific attention markers (e.g., Um hum), narrow medically-focused questions, ignoring 310 patient distress and confusion, and interrupting the patient. Facilitation was covered by showing how provider 311 utterances can effectively elicit patients’ accounts and reinforce question asking. Providers role played both types of 312 providers and reflected on the experience. To increase the effectiveness of the intervention, many of the exercises 313 during the workshop were based on recordings of two audio taped sessions conducted by the provider prior to the 314 training (after securing patient consent). These audio recordings were reviewed by coaches with relevant sections 315 transcribed to provide feedback in the next training session. Reviews of audio-recordings supported the development 316 of attentiveness, facilitation and collaboration in patient encounters. By specifying features of the verbal interaction 317 in the transcription that distinguish between provider “successes” and “challenges” in responsive care, the providers 318 identified how to responsively ask, listen and collaborate. Providers also recognized how patient disclosures lead to 319 exploration of the conditions and circumstances that contribute to mental health problems, activation, and self-320 management. 321

As part of the training workshop, providers received up to two hours of individual coaching to reinforce their 322 reactions to the idea that their responsiveness to patients can change clinical practice. These coaching sessions were 323 based on coding of the two recorded sessions from the provider’s actual clinical encounters prior to participation in 324 the intervention. Feedback was conducted face to face with the coding coach and was based on written structured 325 feedback that was reviewed with the provider. The goal of the training was to deepen the reflection process and 326 learning through application of skills acquired in real life case material. The emphasis was on giving the provider 327 specific feedback based on an analysis of their SDM, their attributional errors (if they incorrectly assume the 328 patient’s age or education as compared to the actual information for the patient) as well as receptivity of patient’s 329 activation and self-management. Most of the integration of the trainings was done by providing explicit feedback of 330 how they could conduct the interview differently to promote patient activation and self-management as well as 331 SDM. In a small group format utilizing active learning, providers supported each other, discussing the 332 organizational and clinical barriers that might interfere with institutionalizing attentiveness, facilitation and 333 collaboration in care. The goal of the training was to provide individual learning and application of the knowledge 334 acquired in the workshop as well as to promote reflection to raise awareness on the part of providers. 335

Part 2: Follow up individual training session (up to a maximum of 6 hours) took place within two to six months of 336 the training workshop (scheduled at the provider’s convenience and also depending on the time it took to recruit 337 eligible patients and invite them to be in the study and get an appointment with their provider where we could 338 audiotape their session). The goal of these calls (phone) was to discuss any remaining questions regarding the 339 training, using the additional up to 6 audio taped sessions conducted by the provider following the training. Ideally, 340 each provider had 6 participating patients that were recruited and consented to be in the study (half participating in 341 the DECIDE PA and half in the usual care condition), but due to a limited patient pool, providers typically had 342 fewer than 6 participating patients. The goal of the follow up sessions was to identify more topics from the training 343 to review in some detail. The review was meant to give participants more training in areas that were not mastered 344 during the first two sessions, using examples from the tapes and providing more feedback. Structured individual 345 feedback was provided for each recorded session covering the topics in the workshop and specifying areas of 346


strength, areas of average functioning and areas in need for improvement. Written feedback was emailed prior to 347 each call to facilitate discussion and practice of skills taught. 348

Part 3: This consisted of a wrap-up session to summarize and conclude the coaching that providers received. 349

V.A.2. Implementation of Provider Coaching (PC) Intervention. The PC Intervention began with recruiting 350 providers at the designated clinics. Both Patient and Provider DECIDE Interventions were reviewed and received 351 input from our Community Advisory Board (CAB). This allowed for the first phase of participant recruitment and 352 randomization to occur promptly. Participating providers signed a consent form, agreeing to attend the provider 353 trainings and to complete a process interview for 6 of their participating patients during the following 6 months of 354 the project. In addition, a baseline and 6 month follow-up interview were collected. Once the providers had been 355 trained, the implementation phase of the patient intervention began. Providers could opt out of the coaching 356 intervention at any time. 357

V.A.3. Patient Activation Intervention (DECIDE-PA). We conducted three trainings with behavioral health 358 patients who were randomized to the intervention. Trainings each lasted 30-45 minutes. Training 1 (Decisions and 359 Agency) increased participants’ awareness of their role in clinical interactions and encouraged participation and 360 decision making in care. Participants were given an overview of training goals and logistics, and were taught 361 question formulation (Brainstorming) and question-asking strategies. Each participant received a Planner 362 summarizing the intervention sessions. Training 2 (the Who, How and Why of Decisions) taught skills for 363 understanding treatment decisions in terms of the roles, processes, and reasons involved. Care Managers reviewed 364 the practice assignment (i.e. asking questions of providers). Because some patients were critical of their 365 ‘performance’ while others were successful, this was a time where tailoring and addressing barriers was critical. 366 Role-playing and practice assignments helped strengthen learning. Training 3 (Self-Efficacy and Consolidation) was 367 a self-efficacy module, in which participants learned different ways to help answer questions about their behavioral 368 health conditions or treatment options, such as consulting information on evidence-based practices. More time was 369 spent reinforcing the skills learned and identifying areas in need of review 370

VI. Biostatistical Analysis 371

VI. A. Data and Measures: The main outcome was SDM, as measured by the OPTION, a scale based on the blind 372 coded audio recording of the clinical visit at RA2, or what we refer to as the blind coder SDM. OPTION is an 373 observer-rated tool developed from a SDM competencies framework including: problem definition, through 374 exploration and discussion of alternative choices, explanation of options and risk and engagement in the decision 375 making process. The OPTION scale measures overall provider involvement in these competencies and provides an 376 indication of the quality of provider involvement in the SDM process as well as precise areas for improvement. 377 Following Legare and colleagues’ recommendation42 of the need to triangulate SDM measurement from multiple 378 perspectives (patient, provider and independent observer) we also use the SDM-9 to assess patient-reported SDM43 ( 379 which we refer to in the final report as patient’s SDM) from the visit and separately the SDM-Q-9 Doc44 to get the 380 provider-reported SDM for that same visit (which we refer to as provider’s SDM). The OPTION (independent 381 observer) is an objective measure of SDM, while the SDM-9 and SDM-Q-9 are subjective measures (i.e. the 382 provider’s and patient’s perception). Representative items include, “My provider wanted to know exactly how I 383 want to be involved in making the decision” (patient), “I told my patient that there are different options for treating 384 his/her condition” (provider). The SDM-9 (patient and provider forms) measures nine constructs (e.g. preferences 385 for involvement, negotiation) deemed essential to SDM. The nine-items are rated on six point scale from 386 “completely disagree” to “completely agree.” A summation of all items results in a raw score of between 0 and 45, 387 transformed by multiplication to result in a range from 0 to 100, where 0 indicates the lowest level and 100 the 388 highest. The analysis of this primary outcome is the focus at RA3. 389

390 Our secondary outcome is patient’s perception of quality of care, measured with the Perceptions of Care (POC)45 391 survey; a patient self-report questionnaire assessing perception of care and interpersonal experience with a 392 provider(s) during an outpatient visit. An 18-item rating system, it measures continuity of care, provider availability, 393 communication, access to provider, and global evaluation of care. It has been utilized to provide precise and detailed 394 feedback about a patient’s experience in care. We have adapted the scale and response format by using a 4-item 395 Likert-type scale from “never” to “always” for each of 10-items. Total scores are scaled from 0 to 100, with higher 396 scores representing better patient evaluation of care. Representative items include, “Does the provider give you 397 reassurance and support?” 398


399 All of the variables and measures collected are listed in Table 1. This includes variables for both patients and 400 providers and at which point in time they were collected. 401

402 Table 1: Description of Outcome Measures, Mediators, and Control Variables 403

Type of Measure Measure Description

Administ

ration

(Baseline

(T1),

Clinical

Visit

(T2),

Post

(T3))

Measure Description

Administration

(Baseline (T1),

Clinical Visit (T2),

Post (T3))

Patient Assessment T1 T2 T3 Provider Assessment T1 T2 T3

Primary Outcome Variables

Self-perceived

Shared Decision

Making (SDM)

Questionnaire:

SDM-Q-943, 44

SDM-Q Patient version: 9-item

scale rates patient perception of

collaboration with provider.

X X X SDM-Q Clinician

Version: 9-item

measure of perceived

SDM; α= .88

X X X

Shared Decision

Making Coded from

Visit: Shared

Decision Making

OPTION

Observer-rated tool of patient

involvement developed from a

framework of SDM

competencies.

X Measures quality of

provider involvement.

X

Shared Decision

Making (internal

measure)

An additional 10-item measure,

developed internally, will also be

used to assess shared decision

making. This measure is more

specific to the mental health

context and addresses aspects of

shared decision making that are

targeted by both the patient and

provider DECIDE interventions.

X An additional 10-item

measure, developed

internally, will also be

used to assess shared

decision making. This

measure is more

specific to the mental

health context and

addresses aspects of

shared decision making

that are targeted by both

the patient and provider

DECIDE interventions.

X

Secondary Outcome Variable

Perceptions of Care

Survey45

18-item scale, focused on

patients' perception of quality of

care.

X X X

Mediators of the Intervention Effect on SDM

Therapeutic

Alliance

Therapeutic alliance goals, tasks,

and bond. Internal consistency

(α=.98).

X X X Provider assessment of

WAI. 12-item version.

X X X

Patient-Provider

Communication:

Sub-scale of Kim

Alliance Scale

Sub-scale of KAS (11 items)

measures patient-provider

communication. Validity and

X X X Provider perception of

patient provider

communication adapted

X X X


(KAS)46 reliability (α=0.87). from KAS.

Mediators of the Intervention Effect on SDM Operating through the DECIDE-PA Patient Intervention

Patient Activation

Scale47 9 item scale. Good internal

consistency: (α=0.82 Spanish and

α=0.75 English).


patient’s activation.

X

Patient’s Perception

of Self-

Management,

Decision Making

Short-form of PEPPI measures

patients' perceived efficacy and

self-management (α=0.91; 0.85

in Spanish).


patient’s self-

management and

decision making.

X

Patient’s Perception

of Provider’s

Demographic

Factors

Accuracy of patient’s perception

of their provider’s ethnicity,

education, SES, and social

position.

X Provider’s perception of

patient’s ethnicity,

education, SES, position.

X

Mediators of the Intervention Effect on SDM Operating through the DECIDE-PC Provider Intervention

Provider Assessment T1 T2 T3

Patient-Centered

Behavior Coding48 Measure of provider’s facilitating and inhibiting behaviors. X

Active Listening

Observation Scale49 Captures the provider’s effort to unravel patient’s reason for the visit. X

Global Rating50, 51 Three one-minute excerpts rated for tension, interest, warmth, engagement,

linking (of the other), and emotional openness.

X

Four Habits Coding

Scheme52 Items measured on 5-point scale: Habit 1 (Invest in the Beginning); Habit 2

(Elicit Patient’s Perspective): Habit 3 (Demonstrate Empathy): Habit 4 (Invest

in the End).

X

Race/Ethnicity and Language Concordance

Patient Assessment T1 T2 T3 Provider Assessment T1 T2 T3

Race/Ethnicity Census definition of

Race/Ethnicity for patient

X Census definition -

provider

X

Language

Proficiency

Evaluated by single item X Evaluated by 1 item X

Mental Health Status / Health Status Variables – For Adjustment in Regression Models

Clinical Global

Impression –

Improvement (CGI-

I)

7-point scale rates patient’s

change in symptoms relative to

baseline state.

X X X Perception of patient’s

CGI-I.

X

Patient Assessment T1 T2 T3

Depression:

Patient Health

Questionnaire

(PHQ-9)53

PHQ-9: 9 criteria for depression screening; Health professional Dx (k = 0.74;

overall accuracy, 88%; sensitivity, 87%; specificity, 88%).

X X

Generalized Anxiety

Disorder: (GAD-

7)54, 55

GAD-7: 7-item measure; good values of sensitivity (86.8%) and specificity

(93.4%); AUC statistically significant [AUC = 0.957-0.985; p < 0.001].

X X

Trauma: Primary

Care-Post Traumatic

Stress Disorder (PC-

PTSD)56, 57

PC-PTSD screen: Brief 4-item screener for PTSD with good sensitivity (0.77),

specificity (0.85), and efficiency of 0.85.

X X

Physical

Comorbidity:

SF-1258/ WHODAS

259

SF-12148 and WHODAS 2149: Functional Health Status (SF-12): physical health

=.91 and emotional health=.92. WHODAS 2, global measure of disability;

0.89.

X X


Chronic Conditions From NLAAS X X

Alcohol and Drugs:

CAGE – AID60 CAGE-AID screener exhibited sensitivity (0.79) / specificity (0.77) for 1+

‘yes’ responses. sensitivity (0.70) and specificity (0.85) for 2+ ‘yes’ responses.

X X

Other Variables

Patient Assessment T1 T2 T

3

Provider Assessment T1 T2 T

3

Patient’s

Demographics and

Socio-contextual

Factors

Includes gender, education,

employment, income insurance,

perceived social status, literacy,

language, preferences in care.

X Includes gender,

education, income,

training, perceived social

status, language.

X

Patient activation

Measure

13-item patient scale, measures

activation among individuals with

mental health.

X X X X

In Vivo Scale 13-item scale, patient rates to

what extent s/he experienced 13

feelings during visit with provider

(ex., nervous, relieved, etc.)

X 14-item scale, provider

rates to what extent

patient seemed to

experience 14 feelings

during visit. The extra

14th item is “satisfied,

overall.”

X

Questions Related to

Decision Making

Ten questions related to how the

patient and his/her provider

communicated with each other

and made decisions during their

most recent visit.

X X X Provider version includes

10 questions related to

how provider and patient

communicate with each

other and make decisions

during most recent visit.

X X X

Relationship with

Patient

NA 23 items, questions relate

to the typical relationship

provider has with patients

in general.

X X X

Provider Bias Eight questions related to provider

characteristics that may influence

dropping out of care.

X X NA

Decision Making-

Clarity of Provider

10-item assessment, patient rates

perceived communication

between him/her and provider.

For example, patient assesses how

often provider uses clear

language, how often provider

makes sure s/he understands

treatment options.

X NA

Reasons for

Termination

10 items, asks patient to rate how

important each reason was in

deciding to terminate treatment.

X X

Perceptions of

Provider/ Patient-

Provider Interaction

“Perceptions of Provider”; 4-item,

patients assesses perceived

race/ethnicity, education and

socioeconomic status of provider.

X X X Provider assesses

perceived race/ethnicity,

education and

socioeconomic status of

patient.

X

Language

Proficiency

4 questions that ask patient how

well s/he is able to read, write and

speak in English. Language

preference for interview is asked.

X 4 questions that ask

providers how well s/he is

able to read, write and

speak in English.

X


404

If second language spoken, they

are asked about that as well.

Language preference for

interview is asked. If

second language is

spoken, they are asked

about that as well.

Services 6-item assessment of service use,

used to assess the patient’s use of

mental health services, such as

therapy or psychological

counseling, reasons for the

services use, and frequency of

services use.

X NA

Health Literacy 18-item assessment. Patient sees a

card with one word on top and

two words on bottom. Patient is

asked to read the top word and to

indicate which of bottom two

words is associated.

X NA

Cultural Training NA One item, Provider is

asked if s/he has received

cultural training before

and how much.

X

Provider’s

Perception

NA 5 items, asks provider to

rate different statements

regarding perceived

patient’s confidence in

addressing concerns

during visit.

X

Barriers to

Treatment

15 items, assesses whether low

perceived need, attitudinal

problems and structural barriers

have prevented patient from

getting help in the last year.

X NA

Questions about

Stigma

2 items, asks patient whether s/he

has been affected by stereotypes

and if s/he has been mistreated

due to her/his identity etc.

X NA

Everyday

Discrimination

2-items taken from the 9-item

Everyday Discrimination Scale.

Patient is asked how often s/he is

treated with less respect than

others and if so, to specify the

reason.

X NA

Preference for

Receiving Patient

Intervention

8 questions, asks patient about

participation in trainings, to

comment on skills they learned

and to comment on anything they

would want.

X NA

Questions about

Shared Decision

Making

2 questions that ask the patient if

s/he has heard of the term “Shared

Decision Making” before and to

elaborate on what s/he thinks the

term means.

X 2 questions that ask if

provider has heard of the

term “Shared Decision

Making” before and to

elaborate on what s/he

thinks the term means.

X


VI. B. Study Endpoints. The study was designed to last 3 years. Data collection concluded in September 2016. 405

VI.C.1. Statistical Methods: Mental Health Status and Health Status Covariates and Other Potential Control 406 Variables: The effect of patient-centered communication has been found to be dependent on patient’s baseline 407 mental health status and illness severity.61

While randomization is likely to provide balance on both unobserved and 408 observed covariates, there is the possibility that one or more of the four study groups may be unbalanced on certain 409 clinical characteristics. As such, in patient- and provider-interviews, we collected additional information at each 410 time period related to mental health status and health status, and we collected demographics at RA1. We compared 411 patients’ demographics across the study groups and found that only personal income differs across groups. The 412 current preliminary analysis does not include covariates except for baseline outcome measures. However in future 413 work, we will run a sensitivity analysis including more covariates. 414

VI.C.2. Analytic Methods: Overview: Table 2 depicts four groups that are analyzed: patients randomized to the 415 DECIDE-PA intervention or usual care were seen by providers who were randomized to DECIDE-PC or no 416 intervention. We describe our analytical approach for each specific aim below. Before formal analyses, we carried 417 out thorough descriptive data analyses to assure that data were free of coding and data entry errors, and to describe 418 the marginal distributions of the key variables. We also described the missing variable patterns and determined 419 whether the patterns varied by design, clinical or demographic features of the participants. To account for missing 420 data, we used multiple imputations which created multiple 421 datasets with missing values replaced by the generated imputed 422 values. Our imputation was done via chained equations which 423 generates predictions based on each conditional density of a 424 variable given other variables.62 Aim 1 and 3 were conducted 425 using imputed data and the non-imputed original data was used 426 for preliminary analysis of Aim 2. We describe the reason for 427 choosing imputed vs. non-imputed data in detail below. 428

VI.C.3. Analysis, Specific Aim 1 (The Effect of DECIDE 429 PA+PC on Shared Decision Making, Patient Activation, Patient’s Perception of Care, Patient Activation): 430 The first aim was to test whether the intervention (DECIDE-PC or DECIDE-PA) had an impact on three outcomes 431 at RA2 and RA3: shared decision making, patient activation, and patient’s perception of quality of their care. We 432 first estimated a multilevel mixed-effects model that allows for random effects at the provider level. Thus, the 433 hierarchical nature of the models accounted for the non-independence of patients seeing the same provider.63 For 434 example, let Yij denotes the RA2 blind coder SDM score for the jth patient who was seen by the ith provider. We 435 estimated the model: 436 437 (1) Yij = 0i + 1(DECIDE-PA)ij + 2(DECIDE-PC)ij + 3(DECIDE-PA+PC)ij + 4Xij k + ij 438

(2) The provider-specific random intercept can be written as: 0i = 00 + *0i 439

where DECIDE-PA was assigned effect codes (-.5, +.5) with -.5 being assigned to patients in the control arm 440 (Category A and B in Table 3) and +.5 being assigned to patients in the treatment arm (Category C and D in Table 441 3). Similarly, DECIDE-PC was effect coded as well where -.5 was assigned to providers in the control groups 442 (Category A and C) and +.5 was assigned to providers in the intervention groups (Category B and D). DECIDE-443 PA+PC denotes the interaction term of the DECIDE-PA and DECIDE-PC. Xij included baseline outcome measures 444 to adjust for imbalance despite random assignment. The term * denotes provider random effects and ij is the 445 individual error term. We ran sensitivity analyses allowing for random effects at the clinic level, but they suggested 446 that random effects at clinic level are minimal, i.e., estimated to be close to zero. 447

Estimations of the model in (1) render the main effects of DECIDE-PA and PC intervention interpretable in the 448 context of interactions. Effect-coded regressions allow us to estimate how much the intervention changed the 449 outcomes across all patients that were affected by the intervention, i.e., the marginal effect of the DECIDE-PA 450 intervention (1), and to estimate how much the intervention changed the outcomes across all provides that received 451 the intervention, i.e., the marginal effect of the DECIDE-PC intervention(2). An interaction term (3) significantly 452 different from zero would suggest additional synergy or anti synergy from the combined DECIDE PA+PC treatment 453 over and above the patient-level intervention DECIDE-PA. We also run estimations of the model in (1) with the 454

Table 2. Randomization Design

Provider Training (PC)

Patient

Training(PA) NO YES

NO

A /Usual

Care B / Only PC

YES C / Only PA D / PA+PC


intervention indicators coded as 0 and 1 (instead of -0.5 and 0.5) but we only report the results of the benchmark 455 regressions as recommended by our consultants. 456

The primary analysis used intent-to-treat principles and assigned all subjects to their randomly determined category, 457 regardless of whether or not they actually received treatment. For patients who didn’t complete RA2 or RA3 458 assessment, we used imputation to project their RA2 and/or RA3 outcomes and accessed them according to the 459 initial random assignment. Therefore, the intent-to-treat 460 analysis used the imputed datasets to fit model in (1). 461

We ran a similar analysis using treatment dosage as the 462 independent variable, where treatment dosage is defined as 463 the number of coaching sessions that patients and/or providers 464 received relative to the intended treatment (up to 3 for 465 patients and up to 6 for providers). While this analysis no 466 longer relies on random assignment, it may still serve as 467 additional confirmation of the results of the intent-to-treat 468 analysis and also provide us with further estimates of the 469 magnitude of the effects. 470

As our primary SDM analysis, we assessed the effect of the 471 intervention on SDM as measured by the blind coder 472 (OPTION scale), a continuous-valued variable standardized to 473 a scale between 0 and 100. As a secondary SDM analysis, we 474 assessed patient- and provider-perceived SDM (SDM-Q-9) 475 scores at RA2 and RA3 using the same model as in (1), 476 controlling for the outcomes at RA1. Two separate models 477 were estimated, one for patient SDM and one for provider 478 SDM. We also adapted the model as in (1) to measure our 479 additional primary outcome variables, patient perceived 480 quality of health care (five continuous variables measured at 481 RA2 and RA3). Further, we ran the same analyses for our 482 secondary outcomes of patient activation (assessed by patient 483 at RA2 and RA3; and by provider only at RA2) as well. 484

VI. C.4. Analysis, Specific Aim 2 (Understanding the Mechanisms by which Enhanced DECIDE Impacts 485 Shared Decision Making and/or Patient Perceived Quality of Care): Aim 2 assessed the mediators for the 486 intervention effects identified in Aim 1. To address not only whether an intervention works but also how it works, 487 we examine as potential mediators: 1) patient-centered communication (KAS) using both audio recordings and 488 patient interview data; 2) therapeutic alliance (WAI) (i.e. the degree to which the patient and provider were 489 “engaged in collaborative, purposive work”);64 and 3) DECIDE-PA outcomes, i.e., patient activation (PAS) and self-490 engagement; perceived efficacy in patient-physician interactions (PEPPI). 491

VI.C.5. Mediation Analysis: The mediation analysis plan built on the analysis described for Aim 1 with the 492 addition of the mediator of interest as a key independent variable. We assessed the impact of the DECIDE-PA and 493 PC intervention on the hypothesized mediators, and then estimated the direct effect of the intervention after 494 adjusting for the mediators. These two analyses allowed the indirect path to be estimated and tested using bootstrap 495 methods or the Sobel test.65, 66 Our bootstrap methods allow for clustering at the provider level and the clinic level. 496 We implemented two different types of models: either a so-called 1-1-1 multi-level mediation model, in which the 497 dependent variable, the independent variable, and the mediator are all at the patient level (level 1); and a 2-1-1 498 multi-level mediation model in which the independent variable is at the provider level (level 2). We used the non-499 imputed data for the preliminary analysis of Aim 2 to circumvent computational constraints imposed by the 500 compound simulation of bootstrap samples and imputed samples. Moreover, we used a preliminary analysis of the a-501 path, i.e., the relationship between the intervention and the mediator, to screen possible mediators. In other words, 502 only mediators affected by interventions can mediate intervention effects, thus, this analysis helps us to narrow the 503 set of potential mediators. Finally, we tested for mediation effect only for candidates suggested by A-path analysis 504 on the intervention effect identified in Aim 1. 505

Figure 2: Study Design


In a causal mediation framework, the identification of mediators is predicated on an unambiguous ordering of the 506 treatment, mediator and outcome variables.67 Having three time periods allowed for an appropriate ordering for the 507 patient- and provider-perceived SDM-Q-9 measure (i.e., randomizationRA2 communication RA3 perceived 508 SDM). We used assessments of the mediator variables at either RA2 or RA3, depending on the timing of the 509 assessment of the outcome variable. If the outcome variable was assessed at RA 2, we only use concurrent mediator 510 variables, i.e., those assessed at RA 2. If the outcome variable was instead assessed at RA 3, we use both concurrent 511 mediators, i.e., those assessed at RA3, and past mediators, i.e., those assessed at RA 2. 512

We recognize the potential bias of concurrent measurement of the mediator and outcome. A possible prospective 513 analysis would minimize this bias using adjustment methods proposed by Shrout et al.68 that account for the 514 information already available in the baseline association between the mediator (i.e., communication) and outcome 515 (perceived SDM). Considering the baseline mediator (measured at RA1) to be an instrumental variable that has an 516 impact on SDM only through its connection with the RA2 mediator, we could identify a residual correlation 517 between the mediator and the outcome variable. Furthermore, we could incorporate the association between the 518 DECIDE intervention and the mechanisms of interest (communication, therapeutic alliance, perceived self-efficacy 519 in patient-provider interactions, etc.) as well as the association between the mechanism and the outcomes, 520 conditional on other covariates. This test of mediation improves upon tests typically conducted69 in that it uses 521 models with a full set of covariates, so models are not prone to omitted variable bias.70 522

VI.C.6. Analyses, Specific Aim 3 (Understanding the Moderating Effects of Racial/Ethnic and Linguistic 523 Discordance on the Impact of DECIDE on Shared Decision Making and/or Patient Perceived Quality of 524 Care): In Aim 3, we explored whether patient/provider racial/ethnic or linguistic discordance moderated the effect 525 of the DECIDE PA and/or PC intervention. We hypothesized that the intervention effect would be greater among 526 discordant dyads because there would be more potential to break down communication and decision-making 527 barriers. Built upon the model described in Aim 1, regression models of Aim 3 included the main effects of patient 528 and provider racial/ethnic groups (and in separate analyses, language groups) and additional interaction terms: 529 (Discordant DECIDE-PA), (Discordant DECIDE-PC), (Discordant DECIDE-PA DECIDE-PC). The 530 DECIDE PA and PC effect as well as racial/ethnic (and linguistic) discordant indicators were assigned effect codes 531 (-.5, +.5) to render their main effects interpretable in the context of interactions. We acknowledged major 532 differences existed across sub-ethnicity, language, and culture within the broad ethnic/racial categories of Latino, 533 Black, Asian, and non-Latino White. However, the relatively small sample of the DECIDE PA and DECIDE-PC 534 only allowed us to assess the differences in racial/ethnic groups or language groups rather than more specific 535 differences. 536

The outcome variables of interest are likely to have a non-linear relationship with the independent variables in 537 regression models. In the case of non-linear models, interaction coefficients do not represent the marginal effect of 538 the interaction term.69 Thus, a possible sensitivity analysis would use a predictive margins approach which applies 539 the model coefficients from the model to subsequent counterfactual populations (i.e., DECIDE-PA with concordant 540 dyads, DECIDE-PA with discordant dyads, DECIDE-PC with concordant dyads, DECIDE-PC with discordant 541 dyads, etc.). This technique, named standardized predictions62, 70 or predictive margins,71 has been used in previous 542 health services studies72-74 and would allow us to compare the effect of the intervention among discordant and 543 concordant dyads, after standardization of all other variables. 544

VII. Risks and Discomforts 545

VII.A.1. Complications of Procedures: We did not foresee any potential complications to occur with this 546 intervention. 547

VII. A.2. Psychosocial (non-medical) Risks to Providers: There were no expected serious risks to the provider 548 participants. Minimal risks included potential loss of confidentiality as in all research. We did everything we could 549 to keep all data we collected confidential. We told provider participants that they may experience some discomfort 550 from having sessions be audio recorded and from receiving feedback of their audio recordings or from receiving 551 feedback in emotion recognition exercises. All provider participants were reminded that their participation was 552 always voluntary and that they could decline continued participation at any time. 553

VII.A.3. Psychosocial (non-medical) Risks to Patient Participants: Minimal risks included potential loss of 554 confidentiality as in all research. The DECIDE intervention has been tested in pilot and multi-site randomized 555


controlled settings with no adverse patient reactions. There was some possibility of patient participant discomfort 556 when discussing behavioral health problems and treatments in the course of the assessments and trainings. Patient 557 participants enrolled in the control and intervention arm of the study were asked to disclose information about their 558 appointments and their interaction with their providers, which may have made them uncomfortable or anxious. The 559 informed consent form explicitly stated that patient participation was voluntary and that patients could skip or refuse 560 to answer any items in the assessments. They could stop participation in the study at any point. 561

VII.A.4. Psychosocial (non-medical) Risks to Community Forum and Focus Group Participants: Minimal 562 risks included potential loss of confidentiality as in all research. 563

VIII. Potential Benefits 564

VIII.A.1. Potential Benefits for Providers: The DECIDE PC was designed to help providers improve therapeutic 565 alliance, patient-provider communication, continuance in care, and satisfaction with services for patients in 566 concordant and discordant ethnic/racial dyads in order to improve SDM. The DECIDE PC training for providers 567 consisted of 1.5 days of training which focused on augmenting patient-centered communication and therapeutic 568 alliance as a possible underlying pathway by which SDM could take place. The training also addressed 1) lack of 569 perspective taking; 2) frequent attributional errors that providers make; and 3) decreased receptivity to patient 570 participation and collaboration in decision making. The training included provider coaching totaling 15-20 hours. 571 Providers learned about their own clinical skills and how to enhance them. 572 573 VIII.A.2 Potential Benefits for Patients: Patients who received the DECIDE-PA may have received a better 574 quality of care from their mental health providers. Patients may have found it useful to learn new ways to talk with 575 their health provider about their treatment and that may have improved their overall care. Patients who did not 576 receive the DECIDE-PA intervention might have benefitted from improved communication or involvement with 577 their provider, if the provider was part of the DECIDE-PC intervention. 578

VIII.B. Potential benefits to society: This study also filled a gap for scientifically rigorous research in clinics that 579 ethnic/racial populations depend on to receive mental health care services.75 Research shows that minority patients 580 do not have equal access to high quality care.16, 17 Administrators and providers are often eager to implement 581 interventions but first need strong evidence of improved quality or outcomes in resource-constrained safety net 582 environments.18 The collaborative engagement of patients, clinicians, and administrators helped ensure that 583 DECIDE PA+PC were relevant and met their needs. Further, the study was designed to triangulate data from 584 multiple perspectives (i.e., patient, clinician, and independent observer) to allow for better measurement of SDM 585 and other outcomes. 586

Ensuring quality in behavioral health treatments is a critically important goal, especially so for racial/ethnic 587 minorities given that they receive less behavioral health care76 and experience more severe consequences from 588 behavioral health disorders than non-Latino Whites.77-80 Yet, quality behavioral health care is contingent upon 589 effective communication and strong therapeutic alliance.81, 82 The DECIDE intervention had the potential to impact 590 quality given the centrality of tailoring behavioral provider practices to respond to patient preferences and 591 concerns83 and its strong correlation with perceived quality of care.84 By improving patient-centered communication 592 and forming a strong therapeutic bond, DECIDE could have helped overcome cultural and social differences across 593 patients and providers allowing for quality care that reduces disparities in service delivery.10, 85, 86

594

IX. Monitoring and Quality Assurance 595

IX.A. Data Collection: To improve the security of data collection, we used Dimagi software’s secure server to 596 collect data via CommCare HQ technology. This was installed in a series of secure tablets through which research 597 assessments were collected. Dimagi utilizes a HIPAA compliant, secure, encrypted server that allows for host 598 intrusion and intrusion monitoring system. All technology can only be accessed through secure and password 599 servers. The CommCareHQ application was installed on tablets and these tablets were made available to research 600 assistants serving as interviewers on the PCORI study. Our research assistants had already undergone training by 601 Dimagi. All data transfers to and from the Dimagi server were conducted over industry standard transmission 602 encryption (HTTPS). All access to the cloud infrastructure was protected behind a firewall and required unique VPN 603 access permissions. All data was transferred through channels that are monitored by intrusion monitoring system. 604


IX.B.1. Safety Monitoring: Privacy and Confidentiality: All documents that include PHI were coded so that 605 identifiers (i.e., names, addresses, and telephone numbers) were removed and separated from the research 606 assessments and completed training materials. Research data and notes from participant observations were stored by 607 research staff in a locked file at each of the study sites and the study data was coded. All materials were securely 608 transported from study sites to the central DRU site at MGH, where they continued to be kept under lock and key. 609 Notes and data were uploaded from project laptops to a secure, password protected network maintained by MGH, 610 for transcription and analysis. In addition to paper forms, assessment data was also collected via Dimagi CommCare 611 HQ. All audio-recorded in-depth interviews were also uploaded immediately to the same secure, password 612 protected server maintained at the MGH. No reports were made public using any names or identifying information. 613 Our coded dataset was stored on a secure central server. Only authorized research staff approved by the site 614 Institutional Review Boards had access to the data. 615

PHI will be destroyed according to standard protocols, 7 years after the completion of the study. Patients were told 616 they could withdraw from the study at any point. Information that had been collected up to that point continued to be 617 used unless specified by the patient. 618

Consent forms were kept at each of the recruitment sites under lock and key as was approved. 619

IX.B.2. Safety Monitoring, Ensuring the Safety of Subjects 620

Provider participants: Provider participation was voluntary and information they provided to study staff 621 throughout the course of the study remains confidential. Providers were assured that none of their study data was 622 reported to the Site PI or their supervisor. They could have elected to withdraw from the study at any time. 623

Patient participants: Patients were informed in advance that their responses to a suicidality screener may have 624 required contact with a provider to maintain their safety. This situation was taken very seriously and followed the 625 emergency protocol developed. Patients were also given contact information for the PI and project coordinator to 626 whom they could directly report any concerns or questions about their study participation. 627

Community Forum and Focus Group participants: There were no serious risks to the participants as a result of 628 these forums and groups. There was always at least some risk of loss of confidentiality associated with participating 629 in a community forum or focus group. To help mitigate this risk, we set out as a ground rule of participation that 630 participants kept confidential the views expressed by their fellow participants and that they did not disclose any 631 information to anyone outside of the focus group 632

IX.B.3. Monitoring Plans for Quality Assurance 633

All research team members followed the procedures of confidentiality adhered to by collaborating institutions. 634 Further, all research staff who worked on this project were required to complete training in data confidentiality and 635 security issues and sign a confidentiality agreement prior to working with patients or handling identifying 636 information. The Community Advisory Board (CAB) of this study and all of the Site PIs worked with the DRU’s 637 research team to ensure that the study was monitored from a scientific and ethical standpoint and we held yearly 638 meetings to assess data collection and management. 639

We provided required research materials, documents and technology, such as audio recorders, to facilitate this work. 640 The CMs and RAs were granted remote access to the MGH server, to be able to upload recordings, tracking 641 materials and other information. This information was monitored through quality control checks by the MGH team. 642 Study staff provided regular supervision and oversight to CMs and RAs. Focus groups were also recorded for 643 quality assurance and transcription purposes. 644

IX.C. Outcomes Monitoring 645

The study staff at DRU worked closely with the Site PI, Care Manager (CM) and Research Assistant (RA) at each 646 participating clinic in the set up and ongoing implementation of the study. We provided required research materials, 647 documents and technology, such as audio recorders, to facilitate this work. The CM and RA were granted remote 648 access to the MGH server, to be able to upload recordings, tracking materials and other information. This 649


information was monitored through quality control checks by the MGH team. Study staff provided regular 650 supervision and oversight to CMs and RAs. 651

Fidelity checks were performed on at least 20% of all DECIDE-PA sessions to ensure interventions were delivered 652 accurately and fully. These were also rated in a series of markers for each intervention. We worked to make sure 653 each CM was delivering the intervention at the highest standards. These checks were performed by adherence 654 checkers who were familiar with the intervention and who had adequate clinical background to provide feedback to 655 CMs. 656

Quality control checks were performed on 15% of all assessments to ensure that the assessments were conducted 657 fully and that the data had been entered correctly. Each paper assessment performed was individually checked to 658 ensure it was entered correctly. Each questionnaire performed using the Dimagi technology was also hand-verified. 659 Each RA was also required to do weekly checks to ensure all consent forms and patient data was saved securely. 660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680


681

X. References 682

1. Wang PS, Lane M, Olfson M, Pincus HA, Wells KB, Kessler RC. Twelve-month use of mental health 683 services in the united states: Results from the national comorbidity survey replication. Arch Gen 684 Psychiatry. 2005; 62(6): 629-640. 685

2. Aron L, Honberg R, Duckworth K et al. Grading the states 2009: A report on america’s health care system 686 for adults with serious mental illness. Arlington, VA: National Alliance on Mental Illness; 2009. 687

3. National Institute of Mental Health. National Institute of Mental Health strategic plan for research. 688 Bethesda, MD: National Institute of Mental Health; 2008. 689

4. New Freedom Commission on Mental Health. Achieving the promise: Transforming mental health care in 690 america. Final report. Rockville, MD: DHHS Pub. No. SMA- 03-3832; 2003. 691

5. Swanson KA, Bastani R, Rubenstein LV, Meredith LS, Ford DE. Effect of mental health care and shared 692 decision making in patient satisfaction in a community sample of patients with depression. Med Care Res 693 Rev2007; 64: 416-430. 694

6. U.S. Congressional Budget Office. Selected CBO publications related to health care legistlation, 2009-695 2010. http://www.cbo.gov/ftpdocs/120xx/doc12033/12-23-selectedhealthcarepublications.pdf. 696

7. Office of Applied Studies, SAMHSA, & Department of Health and Human Services. Substance abuse and 697 mental health services administration: Prevalence of substance use among racial & ethnic subgroups in the 698 U.S. http://www.oas.samhsa.gov/NHSDA/ethnic/ethn1006.htm. 699

8. Tucker CM, Herman KC, Ferdinand LA et al. Providing patient-centered culturally sensitive health care: A 700 formative model. The Counseling Psychologist. 2007; 35(5): 679-705. 701

9. Tucker CM, Herman KC, Pederson TR, Higley BP, Montrichard M, Ivery PD. Cultural sensitivity in 702 physician-patient relationships: Perspectives of an ethnically diverse sample of low-income primary care 703 patients. Med Care. 2003; 41: 859-870. 704

10. Cooper L, Roter D, Johnson R, Ford D, Steinwachs D, Powe N. Patient-centered communication, ratings of 705 care, and concordance of patient and physican race. Ann Intern Med. 2003; 139(11): 907-915. 706

11. Cooper AD, Sterling CP, Bacon MP, Bridgeman B. Does action affect perception or memory? Vision Res. 707 62; 2012: 235-240. 708

12. Polo AJ, Alegría M, Sirkin JT (2012). Increasing the engagement of latinos in services through community-709 derived programs: The right question project–mental health. Professional Psychology: Research and 710 Practice, 43(3), 208. 711

13. Kirmayer L, Groleau D, Guzder J, Blake C, Jarvis E. Cultural consultation: A model of mental health 712 service for multicultural societies. Can J Psychiatry. 2003; 48(3): 145. 713

14. Gilmer TP, Kronick RG. Hard times and health insureance: How many americans will be uninsured by 714 2010? Health Aff. 2009; 28(4): 573-577. 715

15. Pippins JR, Alegria M, Haas JS. Association between language proficiency and the quality of primary care 716 among a national sample of insured latinos. Med Care. 2007; 45: 1020-1025. 717

16. DuBard CA, Gizlice A. A language spoken and differences in health status, access to care, and receipt of 718 preventive services among us hispanics. Am J Public Health. 2008; 98: 2021-2028. (62) 719

17. Bauer AM, Alegria M. Impact of patient language proficiency and interpreter service use on the quality of 720 psychiatric care: A systematic review. Psychiatr Serv. 2010; 61(8): 765-773. 721

18. Bauer AM, Chen CN, Alegria M. English language proficiency and mental health service use among latino 722 and asian americans with mental disorders. Med Care. 2010; 48(12): 1097-1104. 723

19. Alegría M, Polo A, Gao S, Santana L et al. Evaluation of a patient activation and empowerment 724 intervention in mental health care. Med Care. 2008; 46(3): 247-256. 725

20. Cortes DE, Mulvaney-Day N, Fortuna L, Reinfeld S, Alegria M. Patient--provider communication: 726 Understanding the role of patient activation for latinos in mental health treatment. Health Educ Behav. 727 2009; 36(1): 138. 728

21. Alegría M, Carson N, Flores M, Li X et al. Activation, self-management, engagement, and retention in 729 behavioral health care: a randomized clinical trial of the DECIDE intervention. JAMA Psychiatry. 730 2014; 71(5): 557-565. 731

22. Ault-Brutus A, Lee C, Singer S, Allen M, Alegría M. Examining Implementation of a Patient Activation 732 and Self-management Intervention Within the Context of an Effectiveness Trial. Adm Policy Ment Health. 733 2014; 41(6): 777-787. 734


http://www.cbo.gov/ftpdocs/120xx/doc12033/12-23-selectedhealthcarepublications.pdf

http://www.oas.samhsa.gov/NHSDA/ethnic/ethn1006.htm

23. Roter D, Stewart M, Putnam S, Lipkin MJ, Stiles W, Inui T. Communication patterns of primary care 735 physicians. JAMA. 1997; 277(4): 350-356. 736

24. Israel BA, Schulz AJ, Parker EA, Becker AB. Review of community-based research: Assessing partnership 737 approaches to improve public health. Annu Rev Public Health.1998; 19(1): 173-202. 738

25. Patel SR, Bakken S, Ruland C. Recent advances in shared decision making for mental health. Curr Opin 739 Psychiatry. 2008; 21(6): 606-612. 740

26. Smedley B, Stith A, Nelson A. Unequal treatment: Confronting racial and ethnic disparities in healthcare. 741 Washington, D.C.: National Academies Press; 2003. 742

27. Bach PB, Pham HH, Schrag D, Tate RC, Hargraves JL. Primary care physicians who treat blacks and 743 whites. N Engl J Med. 2004; 351(6): 575-584. 744

28. Hasnain-Wynia R, Rittner SS. Improving quality and equity in health care by reducing disparities: Chicago, 745 IL: Northwestern University; 2008. 746

29. Hibbard J. Engaging health care consumers to improve the quality of care. Med Care. 2003; 41(1): 747 Supplement I-61 - I-70. 748

30. Alegria M, Nakash O, Lapatin S, et al. How missing information in diagnosis can lead to disparities in the 749 clinical encounter. J Public Health Manag Pract. 2008; 14(6): S26-S35. 750

31. Alegría M, Sribney W, Perez D, Laderman M, Keefe K. The role of patient activation on patient–provider 751 communication and quality of care for us and foreign born latino patients. J Gen Intern Med. 2009; 24(3): 752 534-541. 753

32. Nakash O, Alegria M. Examination of the role of implicit clinical judgments during the mental health 754 intake. Qual Health Res. 2013; 23(5): 645-654. 755

33. Smith RC. Patient-centered interviewing: An evidence-based method. Lippincott Williams & Wilkins; 756 2001. 757

34. Smith RC, Dwamena FC, Grover M, Coffey J, Frankel RM. Behaviorally defined patient-centered 758 communication—a narrative review of the literature. J Gen Intern Med. 2011; 26(2): 185-191. 759

35. Smith RC, Lyles JS, Mettler J, et al.The effectiveness of intensive training for residents in interviewing. A 760 randomized, controlled study. Ann Intern Med. 1998; 128(2): 118. 761

36. Smith RC, Marshall-Dorsey AA, Osborn GG, et al. Evidence-based guidelines for teaching patient-762 centered interviewing. Patient Educ Couns. 2000; 39(1): 27-36. 763

37. Epstein RM, Franks P, Fiscella K, et al. Measuring patient-centered communication in patient–physician 764 consultations: Theoretical and practical issues. Soc Sci Med. 2005; 61(7): 1516-1528. 765

38. Stein T, Kwan J. Thriving in a busy practice: Physician-patient communication training. Eff Clin Pract. 766 1999; 2(2): 63. 767

39. Center for Mental Health Services. Shared decision-making in mental health care. Practice, research, and 768 future directions. Maryland: Substance Abuse and Mental Health Services Administration; 2011. 769

40. Davis MH. The effects of dispositional empathy on emotional reactions and helping: A multidimensional 770 approach. J Pers. 1983; 51(2): 167-184. 771

41. Ross L, Nisbett RE. The person and the situation: Perspectives of social psychology. New York: McGraw-772 Hill; 1991. 773

42. Légaré F, Moher D, Elwyn G, LeBlanc A, Gravel K. Instruments to assess the perception of physicians in 774 the decision-making process of specific clinical encounters: A systematic review. BMC Med Inform Decis 775 Mak. 2007; 7(1): 30. 776

43. Kriston L, Scholl I, Hölzel L, Simon D, Loh A, Härter M. The 9-item shared decision making questionnaire 777 (sdm-q-9). Development and psychometric properties in a primary care sample. Patient Educ Couns. 2010; 778 80(1): 94-99. 779

44. Scholl I, Kriston L, Dirmaier J, Buchholz A, Härter M. Development and psychometric properties of the 780 shared decision making questionnaire–physician version (SDM-Q-doc). Patient Educ Couns. 2012; 88(2): 781 284-290. 782

45. Eisen SV, Wilcox M, Idiculla T, Speredelozzi A, Dickey B. Assessing consumer perceptions of inpatient 783 psychiatric treatment. Jt Comm J Qual Saf. 2002; 28(9): 510-526. 784

46. Kim S, Boren D, Solem S. The Kim alliance scale. Clin Nurs Res. 2001; 10(3): 314-331. 785 47. Green CA, Perrin NA, Polen MR, Leo MC, Hibbard JH, Tusler M. Development of the patient activation 786

measure for mental health. Adm Policy Ment Health. 2010; 37(4): 327-333. 787 48. Zandbelt LC, Smets EMA, Oort FJ, de Haes, HCJM. Coding patient-centered behavior in the medical 788

encounter. Soc Sci Med. 2005; 61: 661-671. 789


49. Fassaert T, van Dulmen S, Schellevis F, Bensing J. Active listening in medical consultations: Development 790 of the active listening observation scale (alos-global). Patient Educ Couns. 2007; 68: 258-264. 791

50. Hall JA, Roter DL, Rand CS. Communication of affect between patient and physician. J Health Soc Behav. 792 1981; 22: 18-30. 793

51. Roter DL, Hall JA, Blanch-Hartigan D, Larson S, Frankel RM. Slicing it thin: New methods for brief 794 sampling analysis using rias-coded medical dialogue. Patient Educ Couns. 2011; 82: 410-419. 795

52. Krupat E, Frankel R, Stein T, Irish J. The four habits coding scheme: Validation of an instrument to assess 796 clinicians’ communication behavior. Patient Educ Couns. 2007; 62: 38-45. 797

53. Kroenke K, Spitzer RL. The PHQ-9: a new depression diagnostic and severity measure. Psychiatric Annals. 798 2002; 32(9): 509-515. 799

54. Spitzer RL, Kroenke K, Williams JBW, Lowe B. A brief measure for assessing generalized anxiety 800 disorder: The GAD-7. Arch Intern Med. 2006; 166(10): 1092-1097. 801

55. Lish JD, Weissman MM, Adams PB, Hoven CW, Bird H. Family psychiatric screening instruments for 802 epidemilogic studies: pilot testing and validation. Psychiatry Res. 1995; 57: 169-180. 803

56. Prins A, Ouimette P, Kimerling R, et al.The primary care PTSD screen (PC-PTSD): development and 804 operating characteristics. Int J Psychiatry Clint Pract. 2004; 9(1): 9-14. 805

57. Van Dam D, Ehring T, Vedel E, Emmelkamp, PM. Validation of the Primary Care Posttraumatic Stress 806 Disorder screening questionnaire (PC-PTSD) in civilian substance use disorder patients. J Subst Abuse 807 Treat. 2010; 39(2): 105-113. 808

58. Ware Jr JE, Kosinski M, Keller SD. A 12-item short-form health survey: construction of scales and 809 preliminary tests of reliabilit and validity. Med Care. 1996; 34(3): 220. 810

59. Utsun TB, Chatterji S, Kostanjsek N, et al. Developing the World Health Organization disability 811 assessment schedule 2.0. Bull World Health Org. 2010; 88(11): 815-823. 812

60. Brown RL, Rounds LA. Conjoint screening questinnaires for alcohol and other drug abuse: criterion 813 validity in a primary care practice. Wis Med J. 1994; 94(3): 135-140. 814

61. Graugaard PKRA, Holgersen K, Finset A. Communicating with alexithymic and non-alexithymic patients: 815 An experimental study of the effect of psychosocial communication and empathy on patient satisfaction. 816 Psychother Psychosom. 2004; 73(2): 92-100. 817

62. StataCorp LP. 2011. Stata statistical software release 11.0 (release. College station, tx: Stata corporation.) 818 63. Raudenbush SW, Bryk AS. Hierarchical linear models: Applications and data analysis methods (Vol. 1). 819

Sage; 2002. 820 64. Baldwin SA, Wampold BE, Imel ZE. Untangling the alliance-outcome correlation: Exploring the relative 821

importance of therapist and patient variability in the alliance. J Consult Clin Psychol. 2007; 75(6): 842-852. 822 65. MacKinnon DP, Lockwood CM, Hoffman JM, West SG, Sheets V. A comparison of methods to test 823

mediation and other intervening variable effects. Psychol Methods. 2002; 7(1): 83. 824 66. Sobel ME Direct and indirect effects in linear structural equation models. Sociological Methods & 825

Research. 1987; 16(1): 155-176. 826 67. Judd CM, Kenny DA. Process analysis estimating mediation in treatment evaluations. Eval Rev. 1981; 827

5(5): 602-619. 828 68. Shrout P, Keyes K, Ornstein K. Integrating causal analysis into psychopathology research. Causality and 829

Psychopathology: Finding the Determinants of Disorders and Their Cues (eds PE Shrout, KM Keyes & K. 830 Ornstein). Oxford University Press; 2011: 3-24. 831

69. Norton EC, Wang H, Ai C. Computing interaction effects and standard errors in logit and probit models. 832 Stata Journal. 2004; 4: 154-167. 833

70. Kronick R, Gilmer T. Insuring low-income adults: Does public coverage crowd out private? Health Aff. 834 2002; 21(1): 225-239. 835

71. Graubard BI, Korn EL. Predictive margins with survey data. Biometrics. 2004; 55(2): 652-659. (117) 836 72. Davern M, Rodin H, Blewett LA, Call KT Are the current population survey uninsurance estimates too 837

high? An examination of the imputation process. Health Serv Res. 2007; 42(5): 2038-2055. 838 73. Blewett LA, Davidson G, Bramlett MD, Rodin H, Messonnier ML. The impact of gaps in health insurance 839

coverage on immunization status for young children. Health Serv Res. 2008; 43: 1619-1636. 840 74. Wells KB, Tang L, Miranda J, Benjamin B, Duan N, Sherbourne CD. The effects of quality improvement 841

for depression in primary care at nine years. Health Serv Res. 2008; 43(6): 1952-1974. 842 75. Pippins J, Alegria M, Haas J. Association between language proficiency and the quality of primary care 843

among a national sample of insured latinos. Med Care. 2007; 45(11): 1020-1025. 844


76. Institute of Medicine & Committee on Health Insurance Status and its Consequences. America's uninsured 845 crisis: Consequences for health and health care. Washington, D.C.: The National Academies Press; 2009. 846 847

77. Buka SL. Disparities in health status and substance use: Ethnicity and socioeconomic factors. Public Health 848 Rep. 2002; 117(Suppl 1): S118-S125. 849

78. Galea S, Ahern J, Tardiff K, et al. Racial/ethnic disparities in overdose mortality trends in new york city, 850 1990–1998. J Urban Health. 2003; 80(2): 201-211. 851

79. Iguchi MY, Fain T, Ramchand RN, Bell J. How criminal system racial disparities may translate into health 852 disparities. J Health Care Poor Underserved. 2005; 16(4): 48-56. 853

80. Schmid AA. Institutional and Behavioral Economics: Journal Entries for Students and 854 Colleagues. Research in the History of Economic Thought and Methodology. 2007; 25(B): 129. 855

81. Wills CE, Holmes-Rovner M. Integrating decision making and mental health interventions research: 856 Research directions. Clinical Psychology. 2006; 13(1): 9-25. 857

82. Hill CE, Corbett CA. A perspective on the history of process and outcome research in counseling 858 psychology. J Couns Psychol. 1983; 40: 3-24. 859

83. Morales, L.S., Cunningham, W.E., Brown, J.A., Liu, H., & Hays, R.D. Are latinos less satisfied with 860 communication by health care providers? J Gen Intern Med. 1999; 14(7): 409-417. 861

84. Johnson RL, Saha S, Arbelaez JJ, Beach MC, Cooper LA. Racial and ethnic differences in patient 862 perceptions of bias and cultural competence in heath care. J Gen Intern Med. 2004; 19: 101-110. 863

85. Cooper LA, Beach MC, Johnson RL, Inui TS. Delving below the surface. Understanding how race and 864 ethnicity influence relationships in health care. J Gen Intern Med. 2006; 21(1): 21-27. 865

86. Rosen DC, Miller AB, Nakash O, Halperin L, Alegria M. Interpersonal complementarity in the mental 866 health intake: A mixed-methods study. J Couns Psychol. 2012; 59(2): 185-196. 867

868

869 870

871


Final Study Protocol and Analytical Plan...1 Final Study Protocol and Analytical Plan 2...

Documents

Transcript of Final Study Protocol and Analytical Plan...1 Final Study Protocol and Analytical Plan 2...